media.transcribe

Transcribe

Turns audio and video into text, with a timestamp on each passage so you can play the recording from that point.

Find it under Media › Transcripts › Transcribe

Settings

Media: The audio or video column to transcribe.
Engine: The transcription engine. Click it to see where it runs and what it supports.
Language: Auto (detect) by default. Set it if the engine gets the language wrong.
Identify speakers: Labels which passages come from which voice.
Exact speaker count: The exact number of speakers, if you know it.
Minimum speakers: The fewest speakers to look for.
Maximum speakers: The most speakers to look for.
Model size: Larger sizes are more accurate and slower. Whisper uses base unless you choose another.
Voice Activity Detection: Skips stretches with no speech. On by default; reduces made-up text in silence.
Context: Names and terms that appear in the recording, to help the engine spell them.
Clean: The engine's own cleanup of the transcript, where it has one.

The transcript, in the column named Transcript.
Timestamped segments, in the column named Timestamped segments. Document view uses these to play the recording from a passage.
The detected language, when the engine reports it.

Split transcript: Turns an existing timed transcript into separate rows.
Download media: Retrieves a recording from a supported link before transcription.

Names, numbers and people talking over each other are the most common errors. Listen before you quote.
Languages, model sizes and speaker options differ by engine.
Speaker labels such as Speaker 1 separate voices; they don't say who is talking.